Example A-SIGNATURES Re: A Way to express Standard multi-argument Method Signatures in Smalltalk Re: [ANSI-Smalltalk] Next STEPs

[Panu Logic]

Paolo Bonzini wrote:
>
>> 2.  A result type written as 'self'
> That should be #self for obvious reasons.  Misunderstanding the 
> scoping of variables (special variables like self included) is one of 
> the most common mistakes in design Smalltalk DSLs.
If you take the example (about  "Message") and embed it  in a   method 
of the
class Message, you'll  see that even 'self' within it compiles correctly.

This is because in (most?) ST-class-browsers  you can evaluate 'self' 
within a method,
and it will refer to the current class.  This makes sense because the 
method is defined
by that class, not by its instance.


Thus, I should have been more precise about the interpretation of 'self'
in "A-Signatures": It refers to the class whose method-signatures OR
whose meta-class signatures we are specifying.

You could expect that in a class-side method 'self' should evaluate to
the meta-class, not the class, but this seems not to be  the case. 
Therefore I believe  it's best to abide by  this convention - because
it  makes it possible to  evaluate these signatures in any standard
browser easily.


I can see a system where the method-signatures of a class could actually be
coded into a method called #SIGNATURES by  the programmer.  Upon saving
such a method, the system could automatically generate a set of unit-tests
that verify the class indeed is in conformance with its declared types.

BTW. For the standards effort, I think it is definitely a good thing if
we can provide a Standard from which a large portion of the unit-tests
can be automatically generated, easily (i.e. no need to write a parser
for that)..  There's still a plenty of 'semantic' tests left to write by 
hand.


Smalltalk doesn't require type-declarations, which is good, but I think 
they
are a great optional tool for generating unit-tests, as well as for 
specifying
the protocols of abstract (and Standard-) classes - plus for programmer-
communication in general. 

The Romans had it hard because they didn't have a language for
expressing ZERO, whereby their language for communicating
any large numbers became ridiculously difficult. We Smalltalkers
have been hindered  for some time by the fact that there hasn't 
been a standardized, succinct language for expressing method-
signatures in Smalltalk.  I can only conjecture that this must be
one of the reasons why there haven't been real convergence
between the different ST-dialects.


>> 4. Each class used within the expressions is part of the standard.
> The biggest problem is that the standard talks about instance 
> protocols and factory protocols (accessed with global variables), not 
> about classes and metaclasses.
If that's the biggest problem,  we're lucky.  Simply change the words we 
use, either way.


Thanks
-Panu Viljamaa